mirror of
https://github.com/GOSTSec/sgminer
synced 2025-02-07 20:44:19 +00:00
core: simplify kernel selection code.
WIP! Use a string instead of a state-machine-ey kernel selection mechanism where kernel names have to be predefined. This should allow just dropping new kernels into dir `kernel` without bloating the code in three other places. Is in dire need of a cleanup, function parameter check, edge case check - all the usual testing. In particular, checking these definitions/keywords: * enum cl_kernels * kname * [c]gpu[s]->kernel (and similar) * memory cleanup after strdup()?.. * chosen_kernel * queue_scrypt_kernel * strbuf * initCl
This commit is contained in:
Noel Maersk
parent
f2934d8afd
commit
1333ed576d
@ -194,46 +194,24 @@ char *set_thread_concurrency(char *arg)
|
||||
return NULL;
|
||||
}
|
||||
|
||||
static enum cl_kernels select_kernel(char *arg)
|
||||
{
|
||||
if (!strcmp(arg, ALEXKARNEW_KERNNAME))
|
||||
return KL_ALEXKARNEW;
|
||||
if (!strcmp(arg, ALEXKAROLD_KERNNAME))
|
||||
return KL_ALEXKAROLD;
|
||||
if (!strcmp(arg, CKOLIVAS_KERNNAME))
|
||||
return KL_CKOLIVAS;
|
||||
if (!strcmp(arg, ZUIKKIS_KERNNAME))
|
||||
return KL_ZUIKKIS;
|
||||
if (!strcmp(arg, PSW_KERNNAME))
|
||||
return KL_PSW;
|
||||
|
||||
return KL_NONE;
|
||||
}
|
||||
|
||||
char *set_kernel(char *arg)
|
||||
{
|
||||
enum cl_kernels kern;
|
||||
int i, device = 0;
|
||||
char *nextptr;
|
||||
int i, device = 0;
|
||||
|
||||
nextptr = strtok(arg, ",");
|
||||
if (nextptr == NULL)
|
||||
return "Invalid parameters for set kernel";
|
||||
kern = select_kernel(nextptr);
|
||||
if (kern == KL_NONE)
|
||||
return "Invalid parameter to set_kernel";
|
||||
gpus[device++].kernel = kern;
|
||||
|
||||
while ((nextptr = strtok(NULL, ",")) != NULL) {
|
||||
kern = select_kernel(nextptr);
|
||||
if (kern == KL_NONE)
|
||||
return "Invalid parameter to set_kernel";
|
||||
gpus[device++].kname = strdup(nextptr);
|
||||
|
||||
gpus[device++].kernel = kern;
|
||||
}
|
||||
while ((nextptr = strtok(NULL, ",")) != NULL)
|
||||
gpus[device++].kname = strdup(nextptr);
|
||||
|
||||
/* If only one kernel name provided, use same for all GPUs. */
|
||||
if (device == 1) {
|
||||
for (i = device; i < MAX_GPUDEVICES; i++)
|
||||
gpus[i].kernel = gpus[0].kernel;
|
||||
gpus[i].kname = strdup(gpus[0].kname);
|
||||
}
|
||||
|
||||
return NULL;
|
||||
@ -1303,27 +1281,8 @@ static bool opencl_thread_prepare(struct thr_info *thr)
|
||||
if (!cgpu->name)
|
||||
cgpu->name = strdup(name);
|
||||
if (!cgpu->kname)
|
||||
{
|
||||
switch (clStates[i]->chosen_kernel) {
|
||||
case KL_ALEXKARNEW:
|
||||
cgpu->kname = ALEXKARNEW_KERNNAME;
|
||||
break;
|
||||
case KL_ALEXKAROLD:
|
||||
cgpu->kname = ALEXKAROLD_KERNNAME;
|
||||
break;
|
||||
case KL_CKOLIVAS:
|
||||
cgpu->kname = CKOLIVAS_KERNNAME;
|
||||
break;
|
||||
case KL_ZUIKKIS:
|
||||
cgpu->kname = ZUIKKIS_KERNNAME;
|
||||
break;
|
||||
case KL_PSW:
|
||||
cgpu->kname = PSW_KERNNAME;
|
||||
break;
|
||||
default:
|
||||
break;
|
||||
}
|
||||
}
|
||||
cgpu->kname = strdup("ckolivas");
|
||||
|
||||
applog(LOG_INFO, "initCl() finished. Found %s", name);
|
||||
cgtime(&now);
|
||||
get_datestamp(cgpu->init, sizeof(cgpu->init), &now);
|
||||
@ -1347,18 +1306,7 @@ static bool opencl_thread_init(struct thr_info *thr)
|
||||
return false;
|
||||
}
|
||||
|
||||
switch (clState->chosen_kernel) {
|
||||
case KL_ALEXKARNEW:
|
||||
case KL_ALEXKAROLD:
|
||||
case KL_CKOLIVAS:
|
||||
case KL_PSW:
|
||||
case KL_ZUIKKIS:
|
||||
thrdata->queue_kernel_parameters = &queue_scrypt_kernel;
|
||||
break;
|
||||
default:
|
||||
applog(LOG_ERR, "Failed to choose kernel in opencl_thread_init");
|
||||
break;
|
||||
}
|
||||
thrdata->queue_kernel_parameters = &queue_scrypt_kernel;
|
||||
|
||||
thrdata->res = (uint32_t *)calloc(buffersize, 1);
|
||||
|
||||
|
||||
3
miner.h
3
miner.h
@ -471,7 +471,7 @@ struct cgpu_info {
|
||||
|
||||
int64_t max_hashes;
|
||||
|
||||
const char *kname;
|
||||
char *kname;
|
||||
bool mapped;
|
||||
int virtual_gpu;
|
||||
int virtual_adl;
|
||||
@ -482,7 +482,6 @@ struct cgpu_info {
|
||||
|
||||
cl_uint vwidth;
|
||||
size_t work_size;
|
||||
enum cl_kernels kernel;
|
||||
cl_ulong max_alloc;
|
||||
|
||||
int opt_lg, lookup_gap;
|
||||
|
||||
47
ocl.c
47
ocl.c
@ -411,13 +411,11 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
|
||||
char binaryfilename[255];
|
||||
char filename[255];
|
||||
char numbuf[16];
|
||||
char strbuf[255];
|
||||
|
||||
if (cgpu->kernel == KL_NONE) {
|
||||
applog(LOG_INFO, "Selecting kernel ckolivas");
|
||||
clState->chosen_kernel = KL_CKOLIVAS;
|
||||
cgpu->kernel = clState->chosen_kernel;
|
||||
} else {
|
||||
clState->chosen_kernel = cgpu->kernel;
|
||||
if (strcmp(cgpu->kname, "") == 0) {
|
||||
applog(LOG_INFO, "No kernel specified, selecting kernel ckolivas");
|
||||
strcpy(cgpu->kname, "ckolivas");
|
||||
}
|
||||
|
||||
/* For some reason 2 vectors is still better even if the card says
|
||||
@ -431,36 +429,13 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
|
||||
/* All available kernels only support vector 1 */
|
||||
cgpu->vwidth = 1;
|
||||
|
||||
switch (clState->chosen_kernel) {
|
||||
case KL_ALEXKARNEW:
|
||||
applog(LOG_WARNING, "Kernel alexkarnew is experimental.");
|
||||
strcpy(filename, ALEXKARNEW_KERNNAME".cl");
|
||||
strcpy(binaryfilename, ALEXKARNEW_KERNNAME);
|
||||
break;
|
||||
case KL_ALEXKAROLD:
|
||||
applog(LOG_WARNING, "Kernel alexkarold is experimental.");
|
||||
strcpy(filename, ALEXKAROLD_KERNNAME".cl");
|
||||
strcpy(binaryfilename, ALEXKAROLD_KERNNAME);
|
||||
break;
|
||||
case KL_CKOLIVAS:
|
||||
strcpy(filename, CKOLIVAS_KERNNAME".cl");
|
||||
strcpy(binaryfilename, CKOLIVAS_KERNNAME);
|
||||
break;
|
||||
case KL_PSW:
|
||||
applog(LOG_WARNING, "Kernel psw is experimental.");
|
||||
strcpy(filename, PSW_KERNNAME".cl");
|
||||
strcpy(binaryfilename, PSW_KERNNAME);
|
||||
break;
|
||||
case KL_ZUIKKIS:
|
||||
applog(LOG_WARNING, "Kernel zuikkis is experimental.");
|
||||
strcpy(filename, ZUIKKIS_KERNNAME".cl");
|
||||
strcpy(binaryfilename, ZUIKKIS_KERNNAME);
|
||||
/* Kernel only supports lookup-gap 2 */
|
||||
cgpu->lookup_gap = 2;
|
||||
break;
|
||||
case KL_NONE: /* Shouldn't happen */
|
||||
break;
|
||||
}
|
||||
sprintf(strbuf, "%s.cl", cgpu->kname);
|
||||
strcpy(filename, strbuf);
|
||||
strcpy(binaryfilename, cgpu->kname);
|
||||
|
||||
/* Kernel zuikkis only supports lookup-gap 2 */
|
||||
if (strcmp(cgpu->kname, "zuikkis") == 0)
|
||||
cgpu->lookup_gap = 2;
|
||||
|
||||
/* Vectors are hard-set to 1 above. */
|
||||
if (likely(cgpu->vwidth))
|
||||
|
||||
20
sgminer.c
20
sgminer.c
@ -4262,25 +4262,7 @@ void write_config(FILE *fcfg)
|
||||
fputs("\",\n\"kernel\" : \"", fcfg);
|
||||
for(i = 0; i < nDevs; i++) {
|
||||
fprintf(fcfg, "%s", i > 0 ? "," : "");
|
||||
switch (gpus[i].kernel) {
|
||||
case KL_NONE: // Shouldn't happen
|
||||
break;
|
||||
case KL_ALEXKARNEW:
|
||||
fprintf(fcfg, ALEXKARNEW_KERNNAME);
|
||||
break;
|
||||
case KL_ALEXKAROLD:
|
||||
fprintf(fcfg, ALEXKAROLD_KERNNAME);
|
||||
break;
|
||||
case KL_CKOLIVAS:
|
||||
fprintf(fcfg, CKOLIVAS_KERNNAME);
|
||||
break;
|
||||
case KL_PSW:
|
||||
fprintf(fcfg, PSW_KERNNAME);
|
||||
break;
|
||||
case KL_ZUIKKIS:
|
||||
fprintf(fcfg, ZUIKKIS_KERNNAME);
|
||||
break;
|
||||
}
|
||||
fprintf(fcfg, "%s", gpus[i].kname);
|
||||
}
|
||||
|
||||
fputs("\",\n\"lookup-gap\" : \"", fcfg);
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user